[PATCH 0/2] media: amphion: handle runtime resume failures

[PATCH 0/2] media: amphion: handle runtime resume failures

[Jiale Yao]

The Amphion core request, resume, and debugfs write paths assume that
pm_runtime_resume_and_get() always succeeds. If runtime resume fails,
these paths access VPU state or hardware while the device is unavailable
and later drop a runtime PM reference that was not acquired.

Check the resume result before taking the dependent path. The core patch
unwinds its lock and avoids VPU access, while the debugfs write handler
returns the error before accessing the core.

Jiale Yao (2):
  media: amphion: handle core runtime resume failures
  media: amphion: handle debug runtime resume failure

 drivers/media/platform/amphion/vpu_core.c | 13 +++++++++++--
 drivers/media/platform/amphion/vpu_dbg.c  |  6 +++++-
 2 files changed, 16 insertions(+), 3 deletions(-)

-- 
2.34.1

[PATCH 1/2] media: amphion: handle core runtime resume failures

[Jiale Yao]

The core request and resume paths continue with VPU state and hardware
access after pm_runtime_resume_and_get() fails. A failed runtime resume
leaves the device unavailable, so those accesses can fail and the later
runtime PM put does not match an acquired reference.

Check the runtime resume result before accessing the VPU. Release the
core lock and return without touching the hardware when resume fails.

Signed-off-by: Jiale Yao <yaojiale02@163.com>
---
 drivers/media/platform/amphion/vpu_core.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/drivers/media/platform/amphion/vpu_core.c b/drivers/media/platform/amphion/vpu_core.c
index ad54f014d8..3d71442807 100644
--- a/drivers/media/platform/amphion/vpu_core.c
+++ b/drivers/media/platform/amphion/vpu_core.c
@@ -372,7 +372,12 @@ struct vpu_core *vpu_request_core(struct vpu_dev *vpu, enum vpu_core_type type)
 		goto exit;
 
 	mutex_lock(&core->lock);
-	pm_runtime_resume_and_get(core->dev);
+	ret = pm_runtime_resume_and_get(core->dev);
+	if (ret < 0) {
+		mutex_unlock(&core->lock);
+		core = NULL;
+		goto exit;
+	}
 
 	if (core->state == VPU_CORE_DEINIT) {
 		if (vpu_iface_get_power_state(core))
@@ -760,7 +765,11 @@ static int __maybe_unused vpu_core_resume(struct device *dev)
 	int ret = 0;
 
 	mutex_lock(&core->lock);
-	pm_runtime_resume_and_get(dev);
+	ret = pm_runtime_resume_and_get(dev);
+	if (ret < 0) {
+		mutex_unlock(&core->lock);
+		return ret;
+	}
 	vpu_core_get_vpu(core);
 
 	if (core->request_count) {
-- 
2.34.1

[PATCH 2/2] media: amphion: handle debug runtime resume failure

[Jiale Yao]

The debugfs core write handler reads the VPU power state after
pm_runtime_resume_and_get() fails. The device may be unavailable after
a failed runtime resume, and the final runtime PM put does not match an
acquired reference.

Return the resume error before taking the core lock or accessing VPU
hardware.

Signed-off-by: Jiale Yao <yaojiale02@163.com>
---
 drivers/media/platform/amphion/vpu_dbg.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/media/platform/amphion/vpu_dbg.c b/drivers/media/platform/amphion/vpu_dbg.c
index 497ae4e8a2..fff4df4f45 100644
--- a/drivers/media/platform/amphion/vpu_dbg.c
+++ b/drivers/media/platform/amphion/vpu_dbg.c
@@ -382,8 +382,12 @@ static ssize_t vpu_dbg_core_write(struct file *file,
 {
 	struct seq_file *s = file->private_data;
 	struct vpu_core *core = s->private;
+	int ret;
+
+	ret = pm_runtime_resume_and_get(core->dev);
+	if (ret < 0)
+		return ret;
 
-	pm_runtime_resume_and_get(core->dev);
 	mutex_lock(&core->lock);
 	if (vpu_iface_get_power_state(core) && !core->request_count) {
 		dev_info(core->dev, "reset\n");
-- 
2.34.1

Re:[PATCH 1/2] media: amphion: handle core runtime resume failures

[jiale yao]

At 2026-10-19 16:27:53, "Jiale Yao" <yaojiale02@163.com> wrote:
>The core request and resume paths continue with VPU state and hardware
>access after pm_runtime_resume_and_get() fails. A failed runtime resume
>leaves the device unavailable, so those accesses can fail and the later
>runtime PM put does not match an acquired reference.
>
>Check the runtime resume result before accessing the VPU. Release the
>core lock and return without touching the hardware when resume fails.
>

Fixes: f445014a2291 ("media: amphion: fix an issue that using pm_runtime_get_sync incorrectly")

>Signed-off-by: Jiale Yao <yaojiale02@163.com>
>---
> drivers/media/platform/amphion/vpu_core.c | 13 +++++++++++--
> 1 file changed, 11 insertions(+), 2 deletions(-)
>
>diff --git a/drivers/media/platform/amphion/vpu_core.c b/drivers/media/platform/amphion/vpu_core.c
>index ad54f014d8..3d71442807 100644
>--- a/drivers/media/platform/amphion/vpu_core.c
>+++ b/drivers/media/platform/amphion/vpu_core.c
>@@ -372,7 +372,12 @@ struct vpu_core *vpu_request_core(struct vpu_dev *vpu, enum vpu_core_type type)
> 		goto exit;
> 
> 	mutex_lock(&core->lock);
>-	pm_runtime_resume_and_get(core->dev);
>+	ret = pm_runtime_resume_and_get(core->dev);
>+	if (ret < 0) {
>+		mutex_unlock(&core->lock);
>+		core = NULL;
>+		goto exit;
>+	}
> 
> 	if (core->state == VPU_CORE_DEINIT) {
> 		if (vpu_iface_get_power_state(core))
>@@ -760,7 +765,11 @@ static int __maybe_unused vpu_core_resume(struct device *dev)
> 	int ret = 0;
> 
> 	mutex_lock(&core->lock);
>-	pm_runtime_resume_and_get(dev);
>+	ret = pm_runtime_resume_and_get(dev);
>+	if (ret < 0) {
>+		mutex_unlock(&core->lock);
>+		return ret;
>+	}
> 	vpu_core_get_vpu(core);
> 
> 	if (core->request_count) {
>-- 
>2.34.1

Re:[PATCH 2/2] media: amphion: handle debug runtime resume failure

[jiale yao]

At 2026-10-19 16:27:54, "Jiale Yao" <yaojiale02@163.com> wrote:
>The debugfs core write handler reads the VPU power state after
>pm_runtime_resume_and_get() fails. The device may be unavailable after
>a failed runtime resume, and the final runtime PM put does not match an
>acquired reference.
>
>Return the resume error before taking the core lock or accessing VPU
>hardware.
>

Fixes: f445014a2291 ("media: amphion: fix an issue that using pm_runtime_get_sync incorrectly")

>Signed-off-by: Jiale Yao <yaojiale02@163.com>
>---
> drivers/media/platform/amphion/vpu_dbg.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
>diff --git a/drivers/media/platform/amphion/vpu_dbg.c b/drivers/media/platform/amphion/vpu_dbg.c
>index 497ae4e8a2..fff4df4f45 100644
>--- a/drivers/media/platform/amphion/vpu_dbg.c
>+++ b/drivers/media/platform/amphion/vpu_dbg.c
>@@ -382,8 +382,12 @@ static ssize_t vpu_dbg_core_write(struct file *file,
> {
> 	struct seq_file *s = file->private_data;
> 	struct vpu_core *core = s->private;
>+	int ret;
>+
>+	ret = pm_runtime_resume_and_get(core->dev);
>+	if (ret < 0)
>+		return ret;
> 
>-	pm_runtime_resume_and_get(core->dev);
> 	mutex_lock(&core->lock);
> 	if (vpu_iface_get_power_state(core) && !core->request_count) {
> 		dev_info(core->dev, "reset\n");
>-- 
>2.34.1

Re: [PATCH 1/2] media: amphion: handle core runtime resume failures

[Ming Qian(OSS)]

On 10/19/2026 4:27 PM, Jiale Yao wrote:
> The core request and resume paths continue with VPU state and hardware
> access after pm_runtime_resume_and_get() fails. A failed runtime resume
> leaves the device unavailable, so those accesses can fail and the later
> runtime PM put does not match an acquired reference.
> 
> Check the runtime resume result before accessing the VPU. Release the
> core lock and return without touching the hardware when resume fails.
> 
> Signed-off-by: Jiale Yao <yaojiale02@163.com>

Reviewed-by: Ming Qian <ming.qian@oss.nxp.com>

> ---
>   drivers/media/platform/amphion/vpu_core.c | 13 +++++++++++--
>   1 file changed, 11 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/media/platform/amphion/vpu_core.c b/drivers/media/platform/amphion/vpu_core.c
> index ad54f014d8..3d71442807 100644
> --- a/drivers/media/platform/amphion/vpu_core.c
> +++ b/drivers/media/platform/amphion/vpu_core.c
> @@ -372,7 +372,12 @@ struct vpu_core *vpu_request_core(struct vpu_dev *vpu, enum vpu_core_type type)
>   		goto exit;
>   
>   	mutex_lock(&core->lock);
> -	pm_runtime_resume_and_get(core->dev);
> +	ret = pm_runtime_resume_and_get(core->dev);
> +	if (ret < 0) {
> +		mutex_unlock(&core->lock);
> +		core = NULL;
> +		goto exit;
> +	}
>   
>   	if (core->state == VPU_CORE_DEINIT) {
>   		if (vpu_iface_get_power_state(core))
> @@ -760,7 +765,11 @@ static int __maybe_unused vpu_core_resume(struct device *dev)
>   	int ret = 0;
>   
>   	mutex_lock(&core->lock);
> -	pm_runtime_resume_and_get(dev);
> +	ret = pm_runtime_resume_and_get(dev);
> +	if (ret < 0) {
> +		mutex_unlock(&core->lock);
> +		return ret;
> +	}
>   	vpu_core_get_vpu(core);
>   
>   	if (core->request_count) {

Re: [PATCH 2/2] media: amphion: handle debug runtime resume failure

[Ming Qian(OSS)]

On 10/19/2026 4:27 PM, Jiale Yao wrote:
> The debugfs core write handler reads the VPU power state after
> pm_runtime_resume_and_get() fails. The device may be unavailable after
> a failed runtime resume, and the final runtime PM put does not match an
> acquired reference.
> 
> Return the resume error before taking the core lock or accessing VPU
> hardware.
> 
> Signed-off-by: Jiale Yao <yaojiale02@163.com>

Reviewed-by: Ming Qian <ming.qian@oss.nxp.com>

> ---
>   drivers/media/platform/amphion/vpu_dbg.c | 6 +++++-
>   1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/media/platform/amphion/vpu_dbg.c b/drivers/media/platform/amphion/vpu_dbg.c
> index 497ae4e8a2..fff4df4f45 100644
> --- a/drivers/media/platform/amphion/vpu_dbg.c
> +++ b/drivers/media/platform/amphion/vpu_dbg.c
> @@ -382,8 +382,12 @@ static ssize_t vpu_dbg_core_write(struct file *file,
>   {
>   	struct seq_file *s = file->private_data;
>   	struct vpu_core *core = s->private;
> +	int ret;
> +
> +	ret = pm_runtime_resume_and_get(core->dev);
> +	if (ret < 0)
> +		return ret;
>   
> -	pm_runtime_resume_and_get(core->dev);
>   	mutex_lock(&core->lock);
>   	if (vpu_iface_get_power_state(core) && !core->request_count) {
>   		dev_info(core->dev, "reset\n");